A domain-agnostic continual multi-task learning model for generalized glucose level and hypoglycemia event prediction

Continuous prediction of blood glucose levels and hypoglycemia events is critical for managing type 1 diabetes mellitus (T1DM), particularly under intensive insulin therapy. Existing models focus on a single task, limiting their practicality and adaptability in automated insulin delivery (AID) systems. To address, a domain-agnostic continual multi-task learning (DA-CMTL) model is proposed to perform both tasks within a unified framework. Trained on simulated datasets via Sim2Real transfer and adapted using elastic weight consolidation, DA-CMTL supports generalization across domains. On public datasets (DiaTrend, OhioT1DM, and ShanghaiT1DM), DA-CMTL achieved a root mean squared error of 14.19 mg/dL, mean absolute error of 10.09 mg/dL, and sensitivity/specificity of 89.28%/94.09% for early hypoglycemia detection. Real-world validation using type 2 diabetes-induced rats demonstrated a reduction in time below range from 3.01–2.58%, supporting reliable integration as a safety layer in AID systems. These results highlight DA-CMTL’s robustness, scalability, and potential to improve safety in AID.